2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2012-2013 Intel Corporation
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
37 #include <sys/fcntl.h>
38 #include <sys/ioccom.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
42 #include <sys/systm.h>
44 #include <dev/pci/pcivar.h>
46 #include <geom/geom.h>
48 #include "nvme_private.h"
50 static void nvme_bio_child_inbed(struct bio *parent, int bio_error);
51 static void nvme_bio_child_done(void *arg,
52 const struct nvme_completion *cpl);
53 static uint32_t nvme_get_num_segments(uint64_t addr, uint64_t size,
55 static void nvme_free_child_bios(int num_bios,
56 struct bio **child_bios);
57 static struct bio ** nvme_allocate_child_bios(int num_bios);
58 static struct bio ** nvme_construct_child_bios(struct bio *bp,
61 static int nvme_ns_split_bio(struct nvme_namespace *ns,
66 nvme_ns_ioctl(struct cdev *cdev, u_long cmd, caddr_t arg, int flag,
69 struct nvme_namespace *ns;
70 struct nvme_controller *ctrlr;
71 struct nvme_pt_command *pt;
79 nvme_ns_test(ns, cmd, arg);
81 case NVME_PASSTHROUGH_CMD:
82 pt = (struct nvme_pt_command *)arg;
83 return (nvme_ctrlr_passthrough_cmd(ctrlr, pt, ns->id,
84 1 /* is_user_buffer */, 0 /* is_admin_cmd */));
86 *(off_t *)arg = (off_t)nvme_ns_get_size(ns);
89 *(u_int *)arg = nvme_ns_get_sector_size(ns);
99 nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused,
105 error = securelevel_gt(td->td_ucred, 0);
111 nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused,
119 nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl)
121 struct bio *bp = arg;
124 * TODO: add more extensive translation of NVMe status codes
125 * to different bio error codes (i.e. EIO, EINVAL, etc.)
127 if (nvme_completion_is_error(cpl)) {
129 bp->bio_flags |= BIO_ERROR;
130 bp->bio_resid = bp->bio_bcount;
138 nvme_ns_strategy(struct bio *bp)
140 struct nvme_namespace *ns;
143 ns = bp->bio_dev->si_drv1;
144 err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done);
148 bp->bio_flags |= BIO_ERROR;
149 bp->bio_resid = bp->bio_bcount;
155 static struct cdevsw nvme_ns_cdevsw = {
156 .d_version = D_VERSION,
159 .d_write = physwrite,
160 .d_open = nvme_ns_open,
161 .d_close = nvme_ns_close,
162 .d_strategy = nvme_ns_strategy,
163 .d_ioctl = nvme_ns_ioctl
167 nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns)
169 return ns->ctrlr->max_xfer_size;
173 nvme_ns_get_sector_size(struct nvme_namespace *ns)
175 return (1 << ns->data.lbaf[ns->data.flbas.format].lbads);
179 nvme_ns_get_num_sectors(struct nvme_namespace *ns)
181 return (ns->data.nsze);
185 nvme_ns_get_size(struct nvme_namespace *ns)
187 return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns));
191 nvme_ns_get_flags(struct nvme_namespace *ns)
197 nvme_ns_get_serial_number(struct nvme_namespace *ns)
199 return ((const char *)ns->ctrlr->cdata.sn);
203 nvme_ns_get_model_number(struct nvme_namespace *ns)
205 return ((const char *)ns->ctrlr->cdata.mn);
208 const struct nvme_namespace_data *
209 nvme_ns_get_data(struct nvme_namespace *ns)
216 nvme_ns_get_stripesize(struct nvme_namespace *ns)
219 return (ns->stripesize);
223 nvme_ns_bio_done(void *arg, const struct nvme_completion *status)
225 struct bio *bp = arg;
226 nvme_cb_fn_t bp_cb_fn;
228 bp_cb_fn = bp->bio_driver1;
231 free(bp->bio_driver2, M_NVME);
233 if (nvme_completion_is_error(status)) {
234 bp->bio_flags |= BIO_ERROR;
235 if (bp->bio_error == 0)
239 if ((bp->bio_flags & BIO_ERROR) == 0)
242 bp->bio_resid = bp->bio_bcount;
244 bp_cb_fn(bp, status);
248 nvme_bio_child_inbed(struct bio *parent, int bio_error)
250 struct nvme_completion parent_cpl;
253 if (bio_error != 0) {
254 parent->bio_flags |= BIO_ERROR;
255 parent->bio_error = bio_error;
259 * atomic_fetchadd will return value before adding 1, so we still
260 * must add 1 to get the updated inbed number. Save bio_children
261 * before incrementing to guard against race conditions when
262 * two children bios complete on different queues.
264 children = atomic_load_acq_int(&parent->bio_children);
265 inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1;
266 if (inbed == children) {
267 bzero(&parent_cpl, sizeof(parent_cpl));
268 if (parent->bio_flags & BIO_ERROR)
269 parent_cpl.status.sc = NVME_SC_DATA_TRANSFER_ERROR;
270 nvme_ns_bio_done(parent, &parent_cpl);
275 nvme_bio_child_done(void *arg, const struct nvme_completion *cpl)
277 struct bio *child = arg;
281 parent = child->bio_parent;
282 g_destroy_bio(child);
283 bio_error = nvme_completion_is_error(cpl) ? EIO : 0;
284 nvme_bio_child_inbed(parent, bio_error);
288 nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align)
290 uint32_t num_segs, offset, remainder;
295 KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n"));
297 num_segs = size / align;
298 remainder = size & (align - 1);
299 offset = addr & (align - 1);
300 if (remainder > 0 || offset > 0)
301 num_segs += 1 + (remainder + offset - 1) / align;
306 nvme_free_child_bios(int num_bios, struct bio **child_bios)
310 for (i = 0; i < num_bios; i++) {
311 if (child_bios[i] != NULL)
312 g_destroy_bio(child_bios[i]);
315 free(child_bios, M_NVME);
319 nvme_allocate_child_bios(int num_bios)
321 struct bio **child_bios;
324 child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT);
325 if (child_bios == NULL)
328 for (i = 0; i < num_bios; i++) {
329 child_bios[i] = g_new_bio();
330 if (child_bios[i] == NULL)
335 nvme_free_child_bios(num_bios, child_bios);
343 nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios)
345 struct bio **child_bios;
351 #ifdef NVME_UNMAPPED_BIO_SUPPORT
356 *num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount,
358 child_bios = nvme_allocate_child_bios(*num_bios);
359 if (child_bios == NULL)
362 bp->bio_children = *num_bios;
364 cur_offset = bp->bio_offset;
365 rem_bcount = bp->bio_bcount;
367 #ifdef NVME_UNMAPPED_BIO_SUPPORT
368 ma_offset = bp->bio_ma_offset;
372 for (i = 0; i < *num_bios; i++) {
373 child = child_bios[i];
374 child->bio_parent = bp;
375 child->bio_cmd = bp->bio_cmd;
376 child->bio_offset = cur_offset;
377 child->bio_bcount = min(rem_bcount,
378 alignment - (cur_offset & (alignment - 1)));
379 child->bio_flags = bp->bio_flags;
380 #ifdef NVME_UNMAPPED_BIO_SUPPORT
381 if (bp->bio_flags & BIO_UNMAPPED) {
382 child->bio_ma_offset = ma_offset;
385 nvme_get_num_segments(child->bio_ma_offset,
386 child->bio_bcount, PAGE_SIZE);
387 ma_offset = (ma_offset + child->bio_bcount) &
389 ma += child->bio_ma_n;
395 child->bio_data = data;
396 data += child->bio_bcount;
398 cur_offset += child->bio_bcount;
399 rem_bcount -= child->bio_bcount;
406 nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp,
410 struct bio **child_bios;
411 int err, i, num_bios;
413 child_bios = nvme_construct_child_bios(bp, alignment, &num_bios);
414 if (child_bios == NULL)
417 for (i = 0; i < num_bios; i++) {
418 child = child_bios[i];
419 err = nvme_ns_bio_process(ns, child, nvme_bio_child_done);
421 nvme_bio_child_inbed(bp, err);
422 g_destroy_bio(child);
426 free(child_bios, M_NVME);
431 nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp,
434 struct nvme_dsm_range *dsm_range;
438 bp->bio_driver1 = cb_fn;
440 if (ns->stripesize > 0 &&
441 (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) {
442 num_bios = nvme_get_num_segments(bp->bio_offset,
443 bp->bio_bcount, ns->stripesize);
445 return (nvme_ns_split_bio(ns, bp, ns->stripesize));
448 switch (bp->bio_cmd) {
450 err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp);
453 err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp);
456 err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp);
460 malloc(sizeof(struct nvme_dsm_range), M_NVME,
463 bp->bio_bcount/nvme_ns_get_sector_size(ns);
464 dsm_range->starting_lba =
465 bp->bio_offset/nvme_ns_get_sector_size(ns);
466 bp->bio_driver2 = dsm_range;
467 err = nvme_ns_cmd_deallocate(ns, dsm_range, 1,
468 nvme_ns_bio_done, bp);
470 free(dsm_range, M_NVME);
481 nvme_ns_construct(struct nvme_namespace *ns, uint32_t id,
482 struct nvme_controller *ctrlr)
484 struct nvme_completion_poll_status status;
491 if (pci_get_devid(ctrlr->dev) == 0x09538086 && ctrlr->cdata.vs[3] != 0)
493 (1 << ctrlr->cdata.vs[3]) * ctrlr->min_page_size;
496 * Namespaces are reconstructed after a controller reset, so check
497 * to make sure we only call mtx_init once on each mtx.
499 * TODO: Move this somewhere where it gets called at controller
500 * construction time, which is not invoked as part of each
503 if (!mtx_initialized(&ns->lock))
504 mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF);
507 nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data,
508 nvme_completion_poll_cb, &status);
509 while (status.done == FALSE)
511 if (nvme_completion_is_error(&status.cpl)) {
512 nvme_printf(ctrlr, "nvme_identify_namespace failed\n");
517 * If the size of is zero, chances are this isn't a valid
518 * namespace (eg one that's not been configured yet). The
519 * standard says the entire id will be zeros, so this is a
520 * cheap way to test for that.
522 if (ns->data.nsze == 0)
526 * Note: format is a 0-based value, so > is appropriate here,
529 if (ns->data.flbas.format > ns->data.nlbaf) {
530 printf("lba format %d exceeds number supported (%d)\n",
531 ns->data.flbas.format, ns->data.nlbaf+1);
535 if (ctrlr->cdata.oncs.dsm)
536 ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED;
538 if (ctrlr->cdata.vwc.present)
539 ns->flags |= NVME_NS_FLUSH_SUPPORTED;
542 * cdev may have already been created, if we are reconstructing the
543 * namespace after a controller-level reset.
545 if (ns->cdev != NULL)
549 * Namespace IDs start at 1, so we need to subtract 1 to create a
550 * correct unit number.
552 unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1;
555 * MAKEDEV_ETERNAL was added in r210923, for cdevs that will never
556 * be destroyed. This avoids refcounting on the cdev object.
557 * That should be OK case here, as long as we're not supporting PCIe
558 * surprise removal nor namespace deletion.
560 #ifdef MAKEDEV_ETERNAL_KLD
561 ns->cdev = make_dev_credf(MAKEDEV_ETERNAL_KLD, &nvme_ns_cdevsw, unit,
562 NULL, UID_ROOT, GID_WHEEL, 0600, "nvme%dns%d",
563 device_get_unit(ctrlr->dev), ns->id);
565 ns->cdev = make_dev_credf(0, &nvme_ns_cdevsw, unit,
566 NULL, UID_ROOT, GID_WHEEL, 0600, "nvme%dns%d",
567 device_get_unit(ctrlr->dev), ns->id);
569 #ifdef NVME_UNMAPPED_BIO_SUPPORT
570 ns->cdev->si_flags |= SI_UNMAPPED;
573 if (ns->cdev != NULL)
574 ns->cdev->si_drv1 = ns;
579 void nvme_ns_destruct(struct nvme_namespace *ns)
582 if (ns->cdev != NULL)
583 destroy_dev(ns->cdev);